Vehicle trajectories collected by a swarm of drones in Helsinki, Finland

This dataset contains highly detailed, georeferenced vehicle trajectories collected by a swarm of drones during three annual data collection campaigns (2023–2025) in the Jätkäsaari peninsula in Helsinki, Finland. 

In total, around 170'000 unique trajectories were captured across 8 different locations. Each trajectory is recorded with high temporal resolution, with vehicle positions sampled at a frequency of 25 or 30 data points per second.

Data Collection

The drone flights were conducted during both the morning and afternoon peak traffic periods on three days in September over three consecutive years. Specifically, the recording dates were 11th, 12th, and 15th September 2023; 16th–18th September 2024; and 17th–19th September 2025.

Each drone was positioned to hover steadily over a predefined location, capturing high-resolution 4K videos at a frame rate of 25-30 frames per second. In 2023, three drones were deployed, while six drones were used in both 2024 and 2025, allowing for an expanded number of recorded locations. A detailed list of all recorded locations, along with the corresponding years and dates, is provided below.

File and data structure 

The dataset is organized into a collection of compressed ZIP archives, with each file corresponding to a specific combination of recording date and scene. The naming convention for these archives follows the format:

Date_SceneID

• Date: recording date in the format YYYY-MM-DD
• SceneID: identifier of the scene (A, B, C...)

Each ZIP archive contains multiple CSV files, where each file represents the vehicle trajectories extracted from a single recorded video. The CSV filenames follow a structured naming convention:

Date_DroneID_SessionID_SceneID-VideoNumber

• Date: recording date (YYYY-MM-DD)
• DroneID: drone identifier (D1, D2, ..., D6)
• SessionID: identifier of the flight session (e.g., AM1, AM2, …, PM1, PM2). “AM” denotes morning sessions, while “PM” refers to afternoon sessions
• SceneID: scene identifier (from the list of locations above)
• VideoNumber: index of the video within the corresponding flight session

For example, the file 2024-09-17_D1_AM1_A-1.csv contains trajectory data from the first video of scene A, recorded on September 17, 2024, by drone D1 during the first morning session.

Data structure

Each csv file is structured into 7 columns and each row represents one datapoint.